LAMP: Learning a Motion Policy to Repeatedly Navigate in an Uncertain Environment

Mobile robots are often tasked with repeatedly navigating through an environment whose traversability changes over time. These may exhibit some hidden structure, which can be learned. Many studies consider reactive algorithms for online planning, however, these do not take advantage of the past executions navigation task future tasks. In this article, we formalize problem minimizing total expected cost to perform multiple start-to-goal tasks on a roadmap by introducing learned planning problem. We propose method that captures information from learn motion policy handle obstacles robot has seen before. LAMP framework, integrates generated existing stack. Finally, extensive set experiments in simulated and real-world environments show proposed outperforms state-of-the-art 10% 40% terms time travel start goal. also evaluate robustness presence localization mapping errors real robot.